Trust & security

Built so a regulated business can say yes to AI.

Guildia answers your customers for you. Safety and control are not a setting you switch on; they are how the product works. It starts as a new hire, earns independence one duty at a time, and keeps hard limits that nothing can change.

1 · Training

Drafts wait for you

Every new duty starts here. The agent writes a draft, you approve, edit or reject it, and you can see how often it gets it right before you trust it with more.

2 · Promoted

Runs on its own, with an undo window

You promote a duty when it has earned it. Even then, outgoing messages wait in a short undo window so a mistake can be caught before it is sent.

Always · Hard limits

No setting can change these

Not the screens, not your plan, not a cleverly worded message. They are enforced in code, and a few sensitive actions, such as email replies, order cancellations and payment requests, always wait for a person.

What it can never do

  • Give clinical or medical advice. A possible emergency gets your own emergency instructions, word for word, and a person is alerted right away.
  • Message someone who has said STOP. It stops at once, cancels anything queued for them, and does not write again unless they agree again.
  • Message someone without consent on file, or send texts outside quiet hours (8:00 to 20:00 in your local time, or narrower if you choose).
  • Place collections calls or send pressure about money owed. These are not allowed at any level of independence.
  • Pretend to be a person. It says it is an AI when asked and offers a human, and every outgoing call opens by saying so.

What it knows

  • Only what you taught it: answers you approved, from your interview, website and documents. Every reply is written from those approved answers.
  • If there is no approved answer, it says it will check with your team, hands the question to a person, and adds it to a list so you can teach it in one step. It does not guess.

Who sees your data

  • Each business's data is encrypted and kept separate from every other business. It is never sold and never used to train AI models.
  • Customer names and phone numbers stay out of logs and analytics; we work with internal IDs.
  • Access is limited by business and by role. You can export or delete a customer's data, or all of yours, on request.

Health data (HIPAA)

  • For health practices (dental, medical spa, physiotherapy, optometry): no patient names, phone numbers or message bodies in logs, error trackers or analytics.
  • We engineer for least-data handling and pursue Business Associate Agreements with our infrastructure providers. Want the specifics for your compliance review? We will walk you through them.

Audit log

  • Every approval, edit, send and settings change is written to a log that can only be added to. You can always see who did what, and when. Nothing happens off the record.
  • Each business also gets its own trust page, with counts only, which it can choose to make public.

AI disclosure

  • Guildia agents are AI and never hide it. Ask whether it is a bot and it says yes, plainly, and offers to hand you to a person.
  • Outgoing calls open by saying the caller is an AI, and first messages name your business as the sender.

Want to see it before you trust it?

Book a short demo and we will show you the training period, the undo window and the audit log on a real agent.